Skip to main content

koan_server/graphql/
server.rs

1use std::path::PathBuf;
2use std::sync::Arc;
3
4use crossbeam_channel::Sender;
5use koan_core::audio::viz::VizSnapshot;
6use koan_core::auth::{self, parse_duration_secs};
7use koan_core::config::Config;
8use koan_core::db::pool::Pool;
9use koan_core::player::commands::PlayerCommand;
10use koan_core::player::state::SharedPlayerState;
11
12use super::KoanSchema;
13use crate::auth::AuthUser;
14use crate::auth::middleware::{AuthState, auth_middleware};
15use crate::auth::routes::{AuthRouteState, RateLimiter, auth_router};
16
17// ---------------------------------------------------------------------------
18// `koan --headless` entry point (standalone headless server)
19// ---------------------------------------------------------------------------
20
21/// A server hands out links to itself — invites, shares, sign-ins — and is
22/// known to its clients by one address, which it cannot work out for itself.
23fn require_server_url() {
24    let why = match Config::cached().server_url() {
25        None => "server.url is not set. Set it, or KOAN_SERVER__URL, to the address clients reach this server at, e.g. https://koan.example.com".to_owned(),
26        Some(url) if !url.starts_with("https://") && !url.starts_with("http://") => {
27            format!("server.url is {url:?}, which is not an http:// or https:// address")
28        }
29        Some(url) => match crate::base::problem(&url) {
30            Some(why) => format!("server.url is {url:?}: {why}"),
31            None => return,
32        },
33    };
34    log::error!("{why}");
35    eprintln!("koan: {why}");
36    std::process::exit(1);
37}
38
39pub fn cmd_serve(
40    port: Option<u16>,
41    bind: Option<std::net::IpAddr>,
42    subsonic_port: Option<u16>,
43    playground: bool,
44) {
45    use koan_core::player::Player;
46
47    require_server_url();
48    // Open, and migrate, before binding: a database this build cannot read,
49    // such as one a newer koan has migrated, stops it here with the reason
50    // rather than serving errors.
51    if let Err(e) = koan_core::db::connection::Database::open_default() {
52        log::error!("cannot open the database: {e}");
53        eprintln!("koan: cannot open the database: {e}");
54        std::process::exit(1);
55    }
56    let db_path = koan_core::config::db_path();
57    let pool = Arc::new(Pool::new(db_path.clone()));
58    koan_core::db::pool::on_outdated(|| OUTDATED.cancel());
59
60    let (state, _timeline, _viz, cmd_tx) = Player::spawn();
61    // Resolves the library's artist bios and photos ahead of the clients
62    // asking for them.
63    koan_core::artist_info::start();
64
65    // A server has no one to press "scan": index at start and whenever the
66    // library folders change, as the macOS app does.
67    let watched = db_path.clone();
68    let jobs = super::jobs::JobRegistry::default();
69    let record = super::jobs::record_watch(&jobs);
70    koan_core::helpers::spawn_library_watch(db_path, move |scan| {
71        record(&scan);
72        // A scan that just finished may have brought in an album someone asked
73        // to have queued when it arrived.
74        if let koan_core::helpers::WatchScan::Finished(_) = scan {
75            crate::clients::fulfil_from(&watched);
76            if let Ok(db) = koan_core::db::connection::Database::open_existing(&watched) {
77                crate::clients::changed_if_library_moved(&db.conn);
78            }
79        }
80    });
81
82    let started = ApiServer::start_with(
83        ApiServerOpts {
84            state,
85            cmd_tx,
86            pool,
87            port,
88            bind,
89            subsonic_port,
90            playground,
91            viz: None, // headless — no viz analyzer
92            profile: Profile::Headless,
93            hosts: Vec::new(),
94            session: None,
95        },
96        jobs,
97    );
98    match started {
99        Ok(server) => server.wait(),
100        Err(e) => {
101            eprintln!("koan: {}", e);
102            std::process::exit(1);
103        }
104    }
105}
106
107// ---------------------------------------------------------------------------
108// Shared API server logic — used by both headless and TUI+API modes
109// ---------------------------------------------------------------------------
110
111/// Ceiling on a single GraphQL query. Anything longer than this —
112/// a library scan, a remote sync — runs as a job instead.
113const REQUEST_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30);
114
115/// Queries executing at once. Resolvers do their SQLite and HTTP work on the
116/// blocking pool, so this bounds concurrent work rather than protecting the
117/// runtime's workers from it.
118const MAX_INFLIGHT_QUERIES: usize = 64;
119
120/// Largest query body accepted. async-graphql reads the whole body before
121/// parsing it, with no limit of its own. A query is text; even a playlist of
122/// thousands of ids is a fraction of this.
123const MAX_QUERY_BODY: usize = 2 << 20;
124
125/// Timeout, panic catch, body limit and load shed for the query route.
126///
127/// Not applied to `/graphql/ws`: a subscription is meant to outlive any request
128/// timeout.
129fn load_perimeter<S>(router: axum::Router<S>) -> axum::Router<S>
130where
131    S: Clone + Send + Sync + 'static,
132{
133    router
134        // Innermost so it is inside the timeout: a panicking resolver becomes a
135        // 500 rather than a silently dropped connection.
136        .layer(tower_http::catch_panic::CatchPanicLayer::new())
137        .layer(tower_http::limit::RequestBodyLimitLayer::new(
138            MAX_QUERY_BODY,
139        ))
140        .layer(tower_http::timeout::TimeoutLayer::with_status_code(
141            axum::http::StatusCode::REQUEST_TIMEOUT,
142            REQUEST_TIMEOUT,
143        ))
144        // Shed rather than queue. A concurrency limit on its own parks callers
145        // on a semaphore, so an overloaded server answers every client slowly
146        // instead of telling the surplus to come back.
147        .layer(
148            tower::ServiceBuilder::new()
149                .layer(axum::error_handling::HandleErrorLayer::new(
150                    |err: tower::BoxError| async move {
151                        if err.is::<tower::load_shed::error::Overloaded>() {
152                            (
153                                axum::http::StatusCode::SERVICE_UNAVAILABLE,
154                                "server at capacity",
155                            )
156                        } else {
157                            (
158                                axum::http::StatusCode::INTERNAL_SERVER_ERROR,
159                                "internal error",
160                            )
161                        }
162                    },
163                ))
164                .load_shed()
165                .concurrency_limit(MAX_INFLIGHT_QUERIES),
166        )
167}
168
169/// Who the server runs for, which decides what it takes from the `[graphql]`,
170/// `[subsonic]` and `[server]` settings.
171#[derive(Clone, Copy, Debug, PartialEq, Eq)]
172pub enum Profile {
173    /// `koan --headless`: a server and nothing else. Its player is heard by
174    /// nobody.
175    Headless,
176    /// Beside the TUI, on the TUI's player.
177    Tui,
178    /// Inside an app sharing its library, on the app's player. Subsonic is
179    /// always on and only accounts sign in to it. The app owns the process, so
180    /// signals are left to it, and the settings a headless deployment uses
181    /// (demo, proxy sign-in, the playground) do not apply.
182    Embedded,
183}
184
185/// Options for the API server — avoids too-many-arguments.
186pub struct ApiServerOpts {
187    pub state: Arc<SharedPlayerState>,
188    pub cmd_tx: Sender<PlayerCommand>,
189    /// Shared by every route that reads the library outside GraphQL, and by
190    /// the MCP listener when there is one.
191    pub pool: Arc<Pool>,
192    pub port: Option<u16>,
193    pub bind: Option<std::net::IpAddr>,
194    pub subsonic_port: Option<u16>,
195    pub playground: bool,
196    pub viz: Option<Arc<VizSnapshot>>,
197    pub profile: Profile,
198    /// Names this server is reached by beyond `graphql.allowed_hosts`, such as
199    /// the Mac's Bonjour name.
200    pub hosts: Vec<String>,
201    /// Whose saved session `savePlaybackState` writes: the files prefix of
202    /// the device whose player this is (`LinkIdentity::files`). None where
203    /// nothing restores one, as on a headless server.
204    pub session: Option<&'static str>,
205}
206
207/// A running API server. Dropping it asks the server to stop and does not
208/// wait for it.
209pub struct ApiServer {
210    addr: std::net::SocketAddr,
211    /// What the app hosting an embedded server administers it with: an API
212    /// key good for the first admin, made for this run and never written down.
213    host_key: Option<String>,
214    stop: tokio_util::sync::CancellationToken,
215    thread: Option<std::thread::JoinHandle<()>>,
216    browser_policy: Arc<BrowserPolicy>,
217}
218
219impl ApiServer {
220    /// Bind the listeners, then serve on a thread of its own.
221    ///
222    /// `Err` means the server refused to start, on a misconfiguration or a port
223    /// it could not bind, and the caller has to surface it — never a silent
224    /// downgrade to an unauthenticated server. Nothing process-wide is touched
225    /// until the main port is bound.
226    ///
227    /// `jobs` is made by the caller, so a library watch started before the
228    /// server records its scans from the first.
229    pub fn start(opts: ApiServerOpts) -> Result<Self, String> {
230        Self::start_with(opts, Default::default())
231    }
232
233    pub(crate) fn start_with(
234        opts: ApiServerOpts,
235        jobs: super::jobs::JobRegistry,
236    ) -> Result<Self, String> {
237        let ApiServerOpts {
238            state,
239            cmd_tx,
240            pool,
241            port,
242            bind,
243            subsonic_port,
244            playground,
245            viz,
246            profile,
247            hosts,
248            session,
249        } = opts;
250        use axum::routing::{get, post};
251
252        let cfg = Config::cached();
253        let embedded = profile == Profile::Embedded;
254        let port = port.unwrap_or(cfg.graphql.port);
255        // The wizard is for a server with no shell to hand, in a container. One
256        // beside a terminal session or inside an app would open its /setup to
257        // any local process with nothing on screen to say so.
258        let setup_wizard = cfg.graphql.setup_wizard && profile == Profile::Headless;
259        let bind = bind.unwrap_or(cfg.graphql.bind);
260        // An app's server has one port, the one it was given.
261        let subsonic_port = if embedded {
262            None
263        } else {
264            subsonic_port.or(cfg.subsonic.port)
265        };
266        let subsonic_enabled = embedded || cfg.subsonic.enabled;
267        let playground_enabled = !embedded && (playground || cfg.graphql.playground);
268        let auth_enabled = cfg.graphql.auth_enabled;
269        let demo = cfg.server.demo;
270
271        if embedded && !auth_enabled {
272            return Err(
273                "graphql.auth_enabled is off, which would make anyone who reaches \
274                 this port an admin"
275                    .into(),
276            );
277        }
278        if demo && embedded {
279            return Err("server.demo is for a headless server".into());
280        }
281        // A demo is one shared account; without sign-in everyone would be admin.
282        if demo && !auth_enabled {
283            return Err("server.demo needs graphql.auth_enabled".into());
284        }
285
286        // Load or generate Ed25519 keypair for JWT signing. A server's first
287        // start makes its own: it is this server's signing key and nothing
288        // else, a fresh one invalidates no token (there can be none yet), and a
289        // server in a container has no terminal to run `koan auth setup` in
290        // before it starts. Accounts are still created deliberately; until one
291        // exists, nothing signs in.
292        let (private_pem, public_pem) = if auth_enabled {
293            let kp = auth::load_or_generate_keypair().map_err(|e| {
294                format!(
295                    "auth_enabled = true but the keypair could not be loaded or created: {}",
296                    e
297                )
298            })?;
299            // An empty or truncated key file would otherwise leave every request an
300            // unauthenticated admin, which is the opposite of what was asked for.
301            if kp.0.is_empty() || kp.1.is_empty() {
302                return Err("auth_enabled = true but the keypair files are empty. \
303                     Run `koan auth regenerate-keys`."
304                    .into());
305            }
306            kp
307        } else {
308            // When auth is disabled, we still need dummy keys for the route state
309            // (routes exist but won't be hit by middleware). Generate if available.
310            auth::load_or_generate_keypair().unwrap_or_default()
311        };
312
313        let access_ttl = parse_duration_secs(&cfg.graphql.access_token_ttl).unwrap_or(900);
314        let refresh_ttl = parse_duration_secs(&cfg.graphql.refresh_token_ttl).unwrap_or(2_592_000);
315
316        // Process-scoped introspection key for playground access. It is a bearer
317        // credential compared verbatim, so it has to be full-entropy random — a
318        // UUID would leak the server start time and cut the guessable space.
319        let introspection_key = if playground_enabled && auth_enabled {
320            Some(Arc::new(auth::random_token().map_err(|e| {
321                format!("failed to generate introspection key: {}", e)
322            })?))
323        } else {
324            None
325        };
326
327        let proxy_auth = if embedded {
328            None
329        } else {
330            crate::ui::ProxyAuth::from_config(
331                &cfg.graphql.proxy_auth_header,
332                &cfg.graphql.proxy_auth_from,
333            )?
334        };
335
336        // A listener on every interface may share its port with another
337        // process's on loopback, which then answers everything sent to
338        // 127.0.0.1: an app's own administration and Funnel among it. Such a
339        // port counts as taken.
340        if bind.is_unspecified() && port != 0 {
341            // `::` takes IPv4 as well, so both loopbacks count.
342            let loopbacks: &[std::net::IpAddr] = match bind {
343                std::net::IpAddr::V4(_) => &[std::net::Ipv4Addr::LOCALHOST.into()],
344                std::net::IpAddr::V6(_) => &[
345                    std::net::Ipv4Addr::LOCALHOST.into(),
346                    std::net::Ipv6Addr::LOCALHOST.into(),
347                ],
348            };
349            for loopback in loopbacks {
350                std::net::TcpListener::bind((*loopback, port))
351                    .map_err(|e| format!("port {port} is already in use on this machine — {e}"))?;
352            }
353        }
354        let gql_listener = std::net::TcpListener::bind((bind, port))
355            .and_then(|l| l.set_nonblocking(true).map(|()| l))
356            .map_err(|e| {
357                format!("failed to bind GraphQL port {port} — {e} (another instance running?)")
358            })?;
359        let addr = gql_listener
360            .local_addr()
361            .map_err(|e| format!("failed to read the GraphQL listener's address — {e}"))?;
362
363        // `--subsonic <port>` set to something other than the GraphQL port adds
364        // a dedicated Subsonic listener.
365        let sub_listener = subsonic_port
366            .filter(|p| subsonic_enabled && *p != port)
367            .and_then(|sub_port| {
368                match std::net::TcpListener::bind((bind, sub_port))
369                    .and_then(|l| l.set_nonblocking(true).map(|()| l))
370                {
371                    Ok(l) => Some(l),
372                    Err(e) => {
373                        log::warn!(
374                            "Dedicated Subsonic port {sub_port} unavailable — {e}. Mounted on GraphQL port only."
375                        );
376                        None
377                    }
378                }
379            });
380
381        let rt = tokio::runtime::Runtime::new()
382            .map_err(|e| format!("could not start the API server's runtime — {e}"))?;
383        let (gql_listener, sub_listener) = {
384            let _in = rt.enter();
385            let gql = tokio::net::TcpListener::from_std(gql_listener)
386                .map_err(|e| format!("failed to listen on {addr} — {e}"))?;
387            let sub = sub_listener.and_then(|l| match tokio::net::TcpListener::from_std(l) {
388                Ok(l) => Some(l),
389                Err(e) => {
390                    log::warn!("Dedicated Subsonic listener unusable — {e}");
391                    None
392                }
393            });
394            (gql, sub)
395        };
396
397        let host_key = embedded
398            .then(auth::random_token)
399            .transpose()
400            .map_err(|e| format!("failed to generate the host key: {e}"))?;
401
402        if demo {
403            let db = pool.get().map_err(|e| format!("demo: {e}"))?;
404            crate::demo::start(&db.conn).map_err(|e| format!("demo: {e}"))?;
405        }
406
407        // What once encrypted account passwords for Subsonic token auth. Nothing
408        // is encrypted with it any more; a copy left on disk is a liability.
409        let _ = std::fs::remove_file(auth::keypair_dir().join("subsonic.key"));
410
411        crate::auth::set_signing_keys(Arc::new(private_pem.clone()), Arc::new(public_pem.clone()));
412
413        let auth_state = AuthState {
414            public_pem: Arc::new(public_pem.clone()),
415            auth_enabled,
416            introspection_key: introspection_key.clone(),
417            pool: pool.clone(),
418        };
419
420        // One for every door a password comes through, so its limits are shared.
421        let users = Arc::new(crate::auth::password::PasswordVerifier::new(pool.clone()));
422        let auth_route_state = AuthRouteState {
423            users: users.clone(),
424            pool: pool.clone(),
425            private_pem: Arc::new(private_pem),
426            public_pem: Arc::new(public_pem),
427            access_ttl_secs: access_ttl,
428            refresh_ttl_secs: refresh_ttl,
429            cookie_secure: cfg.graphql.cookie_secure,
430            login_limiter: Arc::new(RateLimiter::default()),
431        };
432
433        // Plays accounts forward to ListenBrainz; it sleeps while none do.
434        koan_core::scrobbling::start(pool.path().to_path_buf());
435        // Made here, off the runtime: the relay's client is a blocking one,
436        // which cannot be built inside it, and the routes ask for it there.
437        let _ = crate::push::pusher();
438
439        // Ends graceful shutdown, every open session and the MCP streams.
440        let stop = tokio_util::sync::CancellationToken::new();
441        // An app sharing its library serves its own apps; the web UI and MCP
442        // are for a server people reach with a browser or an assistant.
443        let mcp_routes = (!embedded).then(|| {
444            crate::mcp::router(
445                state.clone(),
446                cmd_tx.clone(),
447                auth_state.clone(),
448                cfg.server_url(),
449                profile == Profile::Headless,
450                session,
451                stop.clone(),
452            )
453        });
454        let covers = Arc::new(crate::covers::Covers::in_config_dir());
455        let schema = super::build_schema_sharing(
456            state,
457            cmd_tx,
458            pool.clone(),
459            viz,
460            jobs.clone(),
461            session,
462            Some(covers.clone()),
463        );
464
465        if auth_enabled {
466            log::info!(
467                "Auth enabled (Ed25519 JWT, access TTL {}s, refresh TTL {}s)",
468                access_ttl,
469                refresh_ttl
470            );
471            let no_accounts = pool
472                .get()
473                .ok()
474                .and_then(|db| koan_core::db::queries::auth::has_users(&db.conn).ok())
475                == Some(false);
476            if no_accounts && setup_wizard {
477                log::warn!(
478                    "No accounts yet: whoever opens /setup first becomes the admin. \
479                     Set graphql.setup_wizard = false to make it with `koan auth setup` instead."
480                );
481            }
482        } else {
483            log::info!("Auth disabled — all requests treated as admin");
484        }
485
486        let mut allowed_hosts = cfg.graphql.allowed_hosts.clone();
487        allowed_hosts.extend(hosts);
488        // On a home network every device has a private address; only a tunnel
489        // client on this machine forwards for anyone.
490        let forwarded_from = if embedded {
491            crate::auth::routes::ForwardedFrom::Loopback
492        } else {
493            crate::auth::routes::ForwardedFrom::Private
494        };
495        let browser_policy = Arc::new(BrowserPolicy {
496            origins: cfg.graphql.cors_origins.clone(),
497            hosts: allowed_hosts,
498            ..Default::default()
499        });
500
501        if cfg.graphql.cors_origins.is_empty() {
502            log::info!("CORS: no origins configured — browsers get no cross-origin access");
503        }
504
505        let serving = stop.clone();
506        let serving_key = host_key.clone();
507        let kept_policy = browser_policy.clone();
508        let thread = std::thread::Builder::new()
509            .name("koan-api".into())
510            .spawn(move || {
511                let stop = serving;
512                rt.block_on(async {
513                    let links = tokio_util::task::TaskTracker::new();
514                    if !embedded {
515                        let stop = stop.clone();
516                        tokio::spawn(async move {
517                            shutdown_signal().await;
518                            stop.cancel();
519                        });
520                    }
521
522                    // GraphQL routes — protected by auth middleware.
523                    //
524                    // The query route carries the load perimeter; the WebSocket route does
525                    // not, because a subscription is meant to outlive any request timeout.
526                    let query_route =
527                        load_perimeter(axum::Router::new().route("/graphql", post(graphql_handler)));
528
529                    let gql_app = axum::Router::new()
530                        .merge(query_route)
531                        .route("/graphql/ws", get(graphql_ws_handler))
532                        .layer(axum::Extension(stop.clone()))
533                        .layer(axum::middleware::from_fn_with_state(
534                            auth_state.clone(),
535                            auth_middleware,
536                        ))
537                        // Runs before auth: a rejected request should never reach a
538                        // credential check, let alone execute.
539                        .layer(axum::middleware::from_fn_with_state(
540                            browser_policy.clone(),
541                            browser_guard,
542                        ))
543                        .with_state(schema);
544
545                    // The web UI checks its own session (a page load without one is sent
546                    // to sign in rather than refused), so it sits outside the GraphQL auth
547                    // layer. Built before the auth routes take their state.
548                    // A standing server keeps every album's grid cover ready; an
549                    // app sharing its own library, and the API beside the
550                    // terminal UI, make them as asked.
551                    if profile == Profile::Headless {
552                        covers.warm(pool.clone());
553                    }
554                    let ui_routes = (!embedded).then(|| crate::ui::router(
555                        pool.clone(),
556                        auth_route_state.clone(),
557                        auth_enabled,
558                        setup_wizard,
559                        covers.clone(),
560                        cfg.server_url(),
561                        cfg.mcp.redirect_hosts.clone(),
562                        proxy_auth,
563                        jobs,
564                    )
565                    // A template that fails to render panics; this makes it a 500.
566                    .layer(tower_http::catch_panic::CatchPanicLayer::new()));
567                    let ui_routes = ui_routes.map(|ui| {
568                        if demo {
569                            ui.layer(axum::middleware::from_fn(crate::demo::guard))
570                        } else {
571                            ui
572                        }
573                    });
574
575                    // Auth routes — always accessible (no auth middleware).
576                    let auth_app = auth_router(auth_route_state);
577
578                    // CORS. An empty origin list emits no `Access-Control-Allow-Origin` at
579                    // all; a wildcard would let any web page read this library.
580                    let origins: Vec<axum::http::HeaderValue> = cfg
581                        .graphql
582                        .cors_origins
583                        .iter()
584                        .filter_map(|o| o.parse().ok())
585                        .collect();
586                    let cors = tower_http::cors::CorsLayer::new()
587                        .allow_origin(origins)
588                        .allow_methods([
589                            axum::http::Method::GET,
590                            axum::http::Method::POST,
591                            axum::http::Method::OPTIONS,
592                        ])
593                        .allow_headers([
594                            axum::http::header::AUTHORIZATION,
595                            axum::http::header::CONTENT_TYPE,
596                            axum::http::HeaderName::from_static("x-introspection-key"),
597                        ])
598                        .allow_credentials(true);
599
600                    // Public by design, so outside the auth layers: each route answers for
601                    // one share's own tracks, or the one cover a notification's signed
602                    // link names, and nothing else. The Host guard still applies.
603                    // An app's own server reads the address as each page is
604                    // asked for, since the app changes it as it runs.
605                    let share_url = match embedded {
606                        true => crate::ServerUrl::Live,
607                        false => crate::ServerUrl::Fixed(cfg.server_url()),
608                    };
609                    let edge = axum::middleware::from_fn_with_state(
610                        crate::base::Base::of(&share_url),
611                        crate::base::edge,
612                    );
613                    let share_routes =
614                        crate::share::router(pool.clone(), share_url, covers.clone())
615                            .layer(tower_http::catch_panic::CatchPanicLayer::new());
616                    // Subsonic on the GraphQL port whenever it is enabled: the remote
617                    // TUI bridge builds its stream URL off the GraphQL base. Built once
618                    // and cloned for the dedicated listener, since each build reads the
619                    // config from disk.
620                    // The demo guard goes on Subsonic itself, so a dedicated Subsonic
621                    // port has it too, and on the web UI above.
622                    let subsonic_merged = subsonic_enabled
623                        .then(|| {
624                            crate::subsonic::subsonic_router(
625                                pool,
626                                covers,
627                                users,
628                                profile,
629                                stop.clone(),
630                                links.clone(),
631                                serving_key,
632                            )
633                        })
634                        .map(|sub| {
635                            if demo {
636                                sub.layer(axum::middleware::from_fn(crate::demo::guard))
637                            } else {
638                                sub
639                            }
640                        });
641                    let subsonic_dedicated = subsonic_merged.clone();
642
643                    // An app's own server serves its apps: Subsonic and the
644                    // share pages. GraphQL and its sign-in would let any
645                    // account drive the app's own player.
646                    let mut app = match embedded {
647                        true => share_routes,
648                        false => auth_app.merge(gql_app).merge(share_routes),
649                    };
650                    if let Some(ui) = ui_routes {
651                        app = app.merge(ui);
652                    }
653                    if let Some(mcp) = mcp_routes {
654                        app = app.merge(mcp);
655                    }
656                    if let Some(sub) = subsonic_merged {
657                        app = app.merge(sub);
658                    }
659                    if playground_enabled {
660                        app = app.route(
661                            "/graphql",
662                            get(graphql_playground).with_state((
663                                introspection_key.clone(),
664                                crate::base::path_of(cfg.server_url().as_deref()),
665                            )),
666                        );
667                    }
668                    // Outermost: a request whose `Host` we do not recognise is refused
669                    // before anything else looks at it. Without this a DNS-rebinding page
670                    // reaches the API as same-origin and CORS stops mattering.
671                    let app = app
672                        .layer(cors)
673                        .layer(axum::middleware::from_fn_with_state(
674                            browser_policy.clone(),
675                            host_guard,
676                        ))
677                        .layer(axum::Extension(forwarded_from));
678
679                    log::info!("GraphQL API on http://{addr}/graphql");
680                    if subsonic_enabled {
681                        log::info!("Subsonic REST on http://{addr}/rest/");
682                    }
683                    if playground_enabled {
684                        let playground_url = match introspection_key {
685                            Some(ref key) => {
686                                format!("http://{addr}/graphql?introspection-key={key}")
687                            }
688                            None => format!("http://{addr}/graphql"),
689                        };
690                        log::info!("GraphiQL: {}", playground_url);
691                        // Open browser on macOS/Linux.
692                        #[cfg(target_os = "macos")]
693                        let _ = std::process::Command::new("open").arg(&playground_url).spawn();
694                        #[cfg(target_os = "linux")]
695                        let _ = std::process::Command::new("xdg-open")
696                            .arg(&playground_url)
697                            .spawn();
698                    }
699                    let gql_server = axum::serve(
700                        gql_listener,
701                        axum::ServiceExt::<axum::extract::Request>::into_make_service_with_connect_info::<std::net::SocketAddr>(
702                            tower::Layer::layer(&edge, app),
703                        ),
704                    )
705                    .with_graceful_shutdown(stop.clone().cancelled_owned());
706
707                    if let Some((sub_listener, sub_app)) = sub_listener.zip(subsonic_dedicated) {
708                        if let Ok(a) = sub_listener.local_addr() {
709                            log::info!("Subsonic REST also on http://{a}/rest/ (dedicated port)");
710                        }
711                        // With connection info, as the main listener: the sign-in
712                        // throttle keys on the client's address.
713                        let sub_server = axum::serve(
714                            sub_listener,
715                            axum::ServiceExt::<axum::extract::Request>::into_make_service_with_connect_info::<std::net::SocketAddr>(
716                                tower::Layer::layer(&edge, sub_app),
717                            ),
718                        )
719                        .with_graceful_shutdown(stop.clone().cancelled_owned());
720
721                        tokio::select! {
722                            r = gql_server => { if let Err(e) = r { log::error!("GraphQL server error: {e}"); } },
723                            r = sub_server => { if let Err(e) = r { log::error!("Subsonic server error: {e}"); } },
724                            _ = drain_deadline(&stop) => log::info!("shutting down with connections still open"),
725                        }
726                    } else {
727                        tokio::select! {
728                            r = gql_server => { if let Err(e) = r { log::error!("GraphQL server error: {e}"); } },
729                            _ = drain_deadline(&stop) => log::info!("shutting down with connections still open"),
730                        }
731                    }
732                    // A link is an upgraded connection, which graceful shutdown
733                    // no longer counts: its session is waited for here, so the
734                    // device is unregistered before `stop` returns rather than
735                    // dropped with the runtime part way through.
736                    links.close();
737                    if tokio::time::timeout(DRAIN, links.wait()).await.is_err() {
738                        log::info!("shutting down with links still open");
739                    }
740                });
741                // Blocking work still running (a scan job, a stream's file read)
742                // is left to finish on its own rather than holding up the stop.
743                rt.shutdown_timeout(DRAIN);
744            })
745            .map_err(|e| format!("could not start the API server thread — {e}"))?;
746
747        Ok(Self {
748            addr,
749            host_key,
750            stop,
751            thread: Some(thread),
752            browser_policy: kept_policy,
753        })
754    }
755
756    /// Names this server answers to that move while it runs, such as the
757    /// device's Tailscale name, which exists only while Tailscale is up, and
758    /// a Funnel's address. `names` is asked again when a request names a host
759    /// not otherwise allowed, off the async threads and at most once every
760    /// `RELEARN` however many such requests arrive, and its answer replaces
761    /// the last.
762    pub fn follow_hosts(&self, names: impl Fn() -> Vec<String> + Send + Sync + 'static) {
763        let _ = self.browser_policy.relearn.set(Box::new(names));
764        self.browser_policy.learn();
765    }
766
767    /// See `host_key` on the struct. `None` but for an embedded server.
768    pub fn host_key(&self) -> Option<&str> {
769        self.host_key.as_deref()
770    }
771
772    /// The address the main listener is bound to.
773    pub fn addr(&self) -> std::net::SocketAddr {
774        self.addr
775    }
776
777    /// Stop accepting, close open sessions, and wait for the server to finish:
778    /// at most `DRAIN` for connections, then `DRAIN` for blocking work.
779    pub fn stop(mut self) {
780        self.stop.cancel();
781        self.join();
782    }
783
784    /// Wait until the server stops on its own: on a signal, for the profiles
785    /// that listen for them.
786    pub fn wait(mut self) {
787        self.join();
788    }
789
790    fn join(&mut self) {
791        if let Some(thread) = self.thread.take()
792            && thread.join().is_err()
793        {
794            log::error!("API server thread panicked");
795        }
796    }
797}
798
799impl Drop for ApiServer {
800    fn drop(&mut self) {
801        self.stop.cancel();
802    }
803}
804
805/// Start the API server beside the TUI, blocking until it stops.
806/// Called from a background thread when TUI mode has API enabled. `session`
807/// is whose saved session `savePlaybackState` writes: the caller's, or none.
808#[allow(clippy::too_many_arguments)]
809pub fn start_api_background(
810    state: Arc<SharedPlayerState>,
811    cmd_tx: Sender<PlayerCommand>,
812    db_path: PathBuf,
813    port: Option<u16>,
814    bind: Option<std::net::IpAddr>,
815    subsonic_port: Option<u16>,
816    playground: bool,
817    session: Option<&'static str>,
818) {
819    let started = ApiServer::start(ApiServerOpts {
820        state,
821        cmd_tx,
822        pool: Arc::new(Pool::new(db_path)),
823        port,
824        bind,
825        subsonic_port,
826        playground,
827        viz: None,
828        profile: Profile::Tui,
829        hosts: Vec::new(),
830        session,
831    });
832    match started {
833        Ok(server) => server.wait(),
834        Err(e) => log::error!("API server not started: {}", e),
835    }
836}
837
838// ---------------------------------------------------------------------------
839// Browser perimeter
840// ---------------------------------------------------------------------------
841
842/// What this server will answer to when the caller is a browser.
843///
844/// Two separate questions: which `Host` values name this server (DNS rebinding),
845/// and which `Origin` values may talk to it (CSRF, cross-site WebSockets).
846#[derive(Default)]
847pub(crate) struct BrowserPolicy {
848    origins: Vec<String>,
849    hosts: Vec<String>,
850    /// The names `relearn` last gave, beside `hosts`.
851    learned: parking_lot::RwLock<Vec<String>>,
852    /// See `ApiServer::follow_hosts`.
853    relearn: std::sync::OnceLock<Relearn>,
854    /// When `relearn` was last asked; held while it is, so requests arriving
855    /// together ask once.
856    learned_at: parking_lot::Mutex<Option<std::time::Instant>>,
857}
858
859type Relearn = Box<dyn Fn() -> Vec<String> + Send + Sync>;
860
861/// The least time between two askings of a server's moving names.
862const RELEARN: std::time::Duration = std::time::Duration::from_secs(5);
863
864impl BrowserPolicy {
865    fn host_allowed(&self, host: &str) -> bool {
866        let bare = strip_port(host);
867        let learned = self.learned.read();
868        if self
869            .hosts
870            .iter()
871            .chain(learned.iter())
872            .any(|h| h.eq_ignore_ascii_case(host) || h.eq_ignore_ascii_case(bare))
873        {
874            return true;
875        }
876        // A rebinding attack needs a name it controls; literals and localhost
877        // resolve to this machine by definition.
878        bare.eq_ignore_ascii_case("localhost") || bare.parse::<std::net::IpAddr>().is_ok()
879    }
880
881    /// An origin is allowed if it is configured, or if it is simply this server
882    /// talking to itself — which is what the bundled playground does.
883    fn origin_allowed(&self, origin: &str, host: Option<&str>) -> bool {
884        if self.origins.iter().any(|o| o == origin) {
885            return true;
886        }
887        match (origin.split_once("://"), host) {
888            (Some((_, authority)), Some(host)) => authority.eq_ignore_ascii_case(host),
889            _ => false,
890        }
891    }
892}
893
894impl BrowserPolicy {
895    /// Whether a host refused now may be allowed once the moving names are
896    /// asked again.
897    /// Never waits: while another request is asking (the Tailscale CLI can
898    /// take seconds), this one is refused rather than holding an async
899    /// worker. `learn` itself runs on the blocking pool.
900    fn may_relearn(&self) -> bool {
901        self.relearn.get().is_some()
902            && self
903                .learned_at
904                .try_lock()
905                .is_some_and(|at| at.is_none_or(|at| at.elapsed() >= RELEARN))
906    }
907
908    /// Ask the moving names again, unless they were asked within `RELEARN`.
909    /// Blocking.
910    fn learn(&self) {
911        let Some(relearn) = self.relearn.get() else {
912            return;
913        };
914        let mut at = self.learned_at.lock();
915        if at.is_some_and(|at| at.elapsed() < RELEARN) {
916            return;
917        }
918        *self.learned.write() = relearn();
919        *at = Some(std::time::Instant::now());
920    }
921}
922
923/// `example.com:4000` -> `example.com`, `[::1]:4000` -> `::1`.
924fn strip_port(host: &str) -> &str {
925    if let Some(rest) = host.strip_prefix('[') {
926        return rest.split(']').next().unwrap_or(rest);
927    }
928    match host.rsplit_once(':') {
929        Some((h, port)) if !port.is_empty() && port.bytes().all(|b| b.is_ascii_digit()) => h,
930        _ => host,
931    }
932}
933
934fn header_str(request: &axum::extract::Request, name: axum::http::HeaderName) -> Option<&str> {
935    request.headers().get(name).and_then(|v| v.to_str().ok())
936}
937
938/// Reject requests carrying an unrecognised `Host`.
939async fn host_guard(
940    axum::extract::State(policy): axum::extract::State<Arc<BrowserPolicy>>,
941    request: axum::extract::Request,
942    next: axum::middleware::Next,
943) -> axum::response::Response {
944    use axum::response::IntoResponse;
945
946    // No `Host` at all means no browser: only HTTP/1.0 and raw tooling omit it,
947    // and neither can be steered by an attacker page.
948    let host = header_str(&request, axum::http::header::HOST)
949        .map(str::to_owned)
950        .or_else(|| request.uri().host().map(str::to_owned));
951
952    if let Some(ref host) = host
953        && !policy.host_allowed(host)
954        && !(policy.may_relearn() && {
955            let asked = policy.clone();
956            let _ = tokio::task::spawn_blocking(move || asked.learn()).await;
957            policy.host_allowed(host)
958        })
959    {
960        log::warn!("rejected request for unrecognised Host: {}", host);
961        return (axum::http::StatusCode::FORBIDDEN, "host not allowed").into_response();
962    }
963
964    next.run(request).await
965}
966
967/// Reject cross-site GraphQL traffic.
968///
969/// Two holes, one guard. A WebSocket handshake is exempt from CORS entirely, so
970/// a foreign page can open `/graphql/ws`, have the browser attach the session
971/// cookie, and read every response. And a POST whose content type is
972/// CORS-safelisted (`text/plain`) is sent without a preflight, yet
973/// async-graphql parses it as JSON regardless — so the mutation lands even
974/// though the reply is unreadable.
975async fn browser_guard(
976    axum::extract::State(policy): axum::extract::State<Arc<BrowserPolicy>>,
977    request: axum::extract::Request,
978    next: axum::middleware::Next,
979) -> axum::response::Response {
980    use axum::response::IntoResponse;
981
982    let host = header_str(&request, axum::http::header::HOST).map(str::to_owned);
983    // No `Origin` means a non-browser client, which CSRF cannot reach.
984    if let Some(origin) = header_str(&request, axum::http::header::ORIGIN)
985        && !policy.origin_allowed(origin, host.as_deref())
986    {
987        log::warn!(
988            "rejected GraphQL request from disallowed Origin: {}",
989            origin
990        );
991        return (axum::http::StatusCode::FORBIDDEN, "origin not allowed").into_response();
992    }
993
994    if request.method() == axum::http::Method::POST && !is_graphql_content_type(&request) {
995        return (
996            axum::http::StatusCode::UNSUPPORTED_MEDIA_TYPE,
997            "content type must be application/json or application/graphql",
998        )
999            .into_response();
1000    }
1001
1002    next.run(request).await
1003}
1004
1005fn is_graphql_content_type(request: &axum::extract::Request) -> bool {
1006    header_str(request, axum::http::header::CONTENT_TYPE).is_some_and(|ct| {
1007        let ct = ct.trim().to_ascii_lowercase();
1008        ct.starts_with("application/json") || ct.starts_with("application/graphql")
1009    })
1010}
1011
1012/// Cancelled when the database turns out to have been upgraded by a newer
1013/// koan, as the one replacing this server does when it starts: this build
1014/// can no longer serve it, so it drains and exits as on SIGTERM.
1015static OUTDATED: std::sync::LazyLock<tokio_util::sync::CancellationToken> =
1016    std::sync::LazyLock::new(tokio_util::sync::CancellationToken::new);
1017
1018/// Resolves on SIGINT, SIGTERM where there is one, or the database being
1019/// upgraded past this build. SIGTERM is how a service manager stops a
1020/// process, and as PID 1 in a container an unhandled one is dropped by the
1021/// kernel, leaving the server running until it is killed.
1022async fn shutdown_signal() {
1023    #[cfg(unix)]
1024    {
1025        use tokio::signal::unix::{SignalKind, signal};
1026        let mut terminate = signal(SignalKind::terminate()).expect("failed to listen for SIGTERM");
1027        tokio::select! {
1028            r = tokio::signal::ctrl_c() => r.expect("failed to listen for ctrl+c"),
1029            _ = terminate.recv() => {}
1030            _ = OUTDATED.cancelled() => {}
1031        }
1032    }
1033    #[cfg(not(unix))]
1034    tokio::select! {
1035        r = tokio::signal::ctrl_c() => r.expect("failed to listen for ctrl+c"),
1036        _ = OUTDATED.cancelled() => {}
1037    }
1038}
1039
1040/// How long shutdown waits for open connections before giving up on them.
1041const DRAIN: std::time::Duration = std::time::Duration::from_secs(10);
1042
1043/// Resolves `DRAIN` after the server is told to stop. Graceful shutdown waits for every
1044/// connection to close, and subscription websockets and audio streams never
1045/// close on their own, so without a deadline the wait ends only in a kill.
1046async fn drain_deadline(stop: &tokio_util::sync::CancellationToken) {
1047    stop.cancelled().await;
1048    tokio::time::sleep(DRAIN).await;
1049}
1050
1051async fn graphql_handler(
1052    axum::Extension(user): axum::Extension<AuthUser>,
1053    axum::extract::State(schema): axum::extract::State<KoanSchema>,
1054    req: async_graphql_axum::GraphQLRequest,
1055) -> async_graphql_axum::GraphQLResponse {
1056    let mut request = req.into_inner();
1057    // The auth middleware always injects AuthUser (anonymous_admin when auth is
1058    // disabled, or a real user when auth is enabled). No fallback needed here.
1059    request = request.data(user);
1060    schema.execute(request).await.into()
1061}
1062
1063/// Subscriptions over a WebSocket, as the account the upgrade request
1064/// authenticated. Closed when that account changes or its token lapses (see
1065/// `Lease`), and when the server stops; otherwise, as when auth is off, it
1066/// stays open.
1067async fn graphql_ws_handler(
1068    axum::Extension(user): axum::Extension<AuthUser>,
1069    lease: Option<axum::Extension<crate::auth::Lease>>,
1070    axum::Extension(stop): axum::Extension<tokio_util::sync::CancellationToken>,
1071    axum::extract::State(schema): axum::extract::State<KoanSchema>,
1072    protocol: async_graphql_axum::GraphQLProtocol,
1073    websocket: axum::extract::WebSocketUpgrade,
1074) -> axum::response::Response {
1075    use axum::extract::ws::{CloseFrame, Message, close_code};
1076    use futures_util::{SinkExt, StreamExt};
1077    websocket
1078        .protocols(async_graphql::http::ALL_WEBSOCKET_PROTOCOLS)
1079        .on_upgrade(move |socket| async move {
1080            let (mut sink, stream) = socket.split();
1081            let serve = async_graphql_axum::GraphQLWebSocket::new_with_pair(
1082                &mut sink, stream, schema, protocol,
1083            )
1084            .on_connection_init(move |_| async move {
1085                let mut data = async_graphql::Data::default();
1086                data.insert(user);
1087                Ok(data)
1088            })
1089            .serve();
1090            let ended = async {
1091                match lease {
1092                    Some(axum::Extension(lease)) => lease.ended().await,
1093                    None => std::future::pending().await,
1094                }
1095            };
1096            let close = tokio::select! {
1097                _ = serve => None,
1098                _ = ended => Some((close_code::NORMAL, "sign in again")),
1099                _ = stop.cancelled() => Some((close_code::AWAY, "server stopping")),
1100            };
1101            if let Some((code, reason)) = close {
1102                let _ = sink
1103                    .send(Message::Close(Some(CloseFrame {
1104                        code,
1105                        reason: reason.into(),
1106                    })))
1107                    .await;
1108            }
1109        })
1110}
1111
1112async fn graphql_playground(
1113    axum::extract::Query(params): axum::extract::Query<std::collections::HashMap<String, String>>,
1114    axum::extract::State((key, base)): axum::extract::State<(Option<Arc<String>>, String)>,
1115) -> axum::response::Response {
1116    use axum::response::IntoResponse;
1117
1118    // If an introspection key exists, require it in the URL.
1119    if let Some(ref expected) = key {
1120        let provided = params.get("introspection-key");
1121        if provided.map(|k| k.as_str()) != Some(expected.as_str()) {
1122            return (
1123                axum::http::StatusCode::FORBIDDEN,
1124                "invalid or missing introspection-key",
1125            )
1126                .into_response();
1127        }
1128    }
1129
1130    // Use async-graphql's built-in GraphiQL (self-contained, no CDN). It
1131    // resolves the endpoint against the origin, so the base path is named.
1132    // Inject the introspection key as a default header so all queries are authed.
1133    let endpoint = format!("{base}/graphql");
1134    let mut source = async_graphql::http::GraphiQLSource::build().endpoint(&endpoint);
1135    if let Some(ref k) = key {
1136        source = source.header("X-Introspection-Key", k.as_str());
1137    }
1138
1139    axum::response::Html(source.finish()).into_response()
1140}
1141
1142/// Run the server as a background daemon (fork + detach).
1143pub fn cmd_serve_daemon(
1144    port: Option<u16>,
1145    bind: Option<std::net::IpAddr>,
1146    subsonic_port: Option<u16>,
1147    playground: bool,
1148) {
1149    use std::fs;
1150    use std::process::Command;
1151
1152    require_server_url();
1153
1154    let cfg = Config::cached();
1155    let port_val = port.unwrap_or(cfg.graphql.port);
1156    let bind_val = bind.unwrap_or(cfg.graphql.bind);
1157
1158    let exe = std::env::current_exe().expect("failed to get current exe path");
1159    let mut cmd = Command::new(exe);
1160
1161    cmd.arg("--headless");
1162    cmd.arg("--port").arg(port_val.to_string());
1163    cmd.arg("--bind").arg(bind_val.to_string());
1164    if let Some(sp) = subsonic_port {
1165        cmd.arg("--subsonic").arg(sp.to_string());
1166    }
1167    if playground || cfg.graphql.playground {
1168        cmd.arg("--playground");
1169    }
1170
1171    cmd.stdin(std::process::Stdio::null());
1172    cmd.stdout(std::process::Stdio::null());
1173    cmd.stderr(std::process::Stdio::null());
1174
1175    let mut child = cmd.spawn().expect("failed to spawn daemon process");
1176    let pid = child.id();
1177
1178    let pid_path = koan_core::config::config_dir().join("koan-serve.pid");
1179    fs::write(&pid_path, pid.to_string()).ok();
1180
1181    std::thread::spawn(move || {
1182        let _ = child.wait();
1183    });
1184
1185    eprintln!("kōan daemon started (pid {}) on port {}", pid, port_val);
1186    if let Some(sp) = subsonic_port {
1187        eprintln!("  Subsonic REST on port {}", sp);
1188    }
1189    eprintln!("  PID file: {}", pid_path.display());
1190}
1191
1192// ---------------------------------------------------------------------------
1193// In-process execution (for MCP `graphql` tool)
1194// ---------------------------------------------------------------------------
1195
1196/// Execute a GraphQL query in-process (no HTTP round-trip).
1197///
1198/// There is no credential to check, so the caller states who the query runs as
1199/// and at what role — see `mcp::mcp_role`.
1200pub async fn execute_in_process(
1201    schema: &KoanSchema,
1202    query: &str,
1203    variables: Option<serde_json::Value>,
1204    caller: AuthUser,
1205) -> serde_json::Value {
1206    let mut request = async_graphql::Request::new(query).data(caller);
1207    if let Some(serde_json::Value::Object(map)) = variables {
1208        let mut gql_vars = async_graphql::Variables::default();
1209        for (k, v) in map {
1210            gql_vars.insert(
1211                async_graphql::Name::new(&k),
1212                async_graphql::Value::from_json(v).unwrap_or(async_graphql::Value::Null),
1213            );
1214        }
1215        request = request.variables(gql_vars);
1216    }
1217    let response = schema.execute(request).await;
1218    serde_json::to_value(&response).unwrap_or(serde_json::Value::Null)
1219}
1220
1221// ---------------------------------------------------------------------------
1222// Tests
1223// ---------------------------------------------------------------------------
1224
1225#[cfg(test)]
1226mod tests {
1227    use super::*;
1228    use axum::body::Body;
1229    use axum::http::{Request as HttpRequest, StatusCode};
1230    use axum::routing::{get, post};
1231    use tower::ServiceExt as _;
1232
1233    fn policy() -> Arc<BrowserPolicy> {
1234        Arc::new(BrowserPolicy {
1235            origins: vec!["https://music.example.com".into()],
1236            hosts: vec!["koan.local".into()],
1237            ..Default::default()
1238        })
1239    }
1240
1241    async fn ok() -> &'static str {
1242        "ok"
1243    }
1244
1245    fn routes() -> axum::Router<Arc<BrowserPolicy>> {
1246        axum::Router::new()
1247            .route("/graphql", post(ok).get(ok))
1248            .route("/graphql/ws", get(ok))
1249    }
1250
1251    async fn run_host(req: HttpRequest<Body>) -> StatusCode {
1252        let app = routes()
1253            .layer(axum::middleware::from_fn_with_state(policy(), host_guard))
1254            .with_state(policy());
1255        app.oneshot(req).await.unwrap().status()
1256    }
1257
1258    async fn run_browser(req: HttpRequest<Body>) -> StatusCode {
1259        let app = routes()
1260            .layer(axum::middleware::from_fn_with_state(
1261                policy(),
1262                browser_guard,
1263            ))
1264            .with_state(policy());
1265        app.oneshot(req).await.unwrap().status()
1266    }
1267
1268    fn json_post(uri: &str) -> axum::http::request::Builder {
1269        HttpRequest::post(uri).header(axum::http::header::CONTENT_TYPE, "application/json")
1270    }
1271
1272    // -- Host allowlist (DNS rebinding) --
1273
1274    /// A name that comes up while the server runs, as the Tailscale name
1275    /// does when Tailscale starts, is allowed once the names are asked
1276    /// again, which a request naming it does; another name stays refused.
1277    #[tokio::test]
1278    async fn moving_names_are_learned_from_the_request_that_names_them() {
1279        let up = Arc::new(std::sync::atomic::AtomicBool::new(false));
1280        let policy = Arc::new(BrowserPolicy {
1281            hosts: vec!["koan.local".into()],
1282            ..Default::default()
1283        });
1284        let tailscale = up.clone();
1285        let _ = policy.relearn.set(Box::new(move || {
1286            match tailscale.load(std::sync::atomic::Ordering::Relaxed) {
1287                true => vec!["mac.tailnet.ts.net".into()],
1288                false => Vec::new(),
1289            }
1290        }));
1291        let app = routes()
1292            .layer(axum::middleware::from_fn_with_state(
1293                policy.clone(),
1294                host_guard,
1295            ))
1296            .with_state(policy.clone());
1297        let ask = |host: &str| {
1298            HttpRequest::get("/graphql")
1299                .header(axum::http::header::HOST, host)
1300                .body(Body::empty())
1301                .unwrap()
1302        };
1303
1304        let status = app
1305            .clone()
1306            .oneshot(ask("mac.tailnet.ts.net"))
1307            .await
1308            .unwrap()
1309            .status();
1310        assert_eq!(status, StatusCode::FORBIDDEN, "Tailscale is stopped");
1311        up.store(true, std::sync::atomic::Ordering::Relaxed);
1312        *policy.learned_at.lock() = None;
1313        let status = app
1314            .clone()
1315            .oneshot(ask("mac.tailnet.ts.net"))
1316            .await
1317            .unwrap()
1318            .status();
1319        assert_eq!(status, StatusCode::OK, "learned once Tailscale is up");
1320        let status = app
1321            .clone()
1322            .oneshot(ask("evil.example"))
1323            .await
1324            .unwrap()
1325            .status();
1326        assert_eq!(status, StatusCode::FORBIDDEN);
1327    }
1328
1329    #[test]
1330    fn host_policy_accepts_loopback_literals_and_configured_names() {
1331        let p = policy();
1332        assert!(p.host_allowed("localhost:4000"));
1333        assert!(p.host_allowed("127.0.0.1:4000"));
1334        assert!(p.host_allowed("192.168.1.20:4000"));
1335        assert!(p.host_allowed("[::1]:4000"));
1336        assert!(p.host_allowed("koan.local"));
1337        assert!(p.host_allowed("koan.local:4000"));
1338    }
1339
1340    #[test]
1341    fn host_policy_rejects_attacker_controlled_names() {
1342        let p = policy();
1343        assert!(!p.host_allowed("evil.com"));
1344        assert!(!p.host_allowed("rebind.evil.com:4000"));
1345        assert!(!p.host_allowed("koan.local.evil.com"));
1346    }
1347
1348    #[tokio::test]
1349    async fn host_guard_rejects_foreign_host() {
1350        let req = json_post("/graphql")
1351            .header(axum::http::header::HOST, "rebind.evil.com")
1352            .body(Body::empty())
1353            .unwrap();
1354        assert_eq!(run_host(req).await, StatusCode::FORBIDDEN);
1355    }
1356
1357    #[tokio::test]
1358    async fn host_guard_allows_known_host_and_missing_host() {
1359        let req = json_post("/graphql")
1360            .header(axum::http::header::HOST, "127.0.0.1:4000")
1361            .body(Body::empty())
1362            .unwrap();
1363        assert_eq!(run_host(req).await, StatusCode::OK);
1364
1365        let req = json_post("/graphql").body(Body::empty()).unwrap();
1366        assert_eq!(run_host(req).await, StatusCode::OK);
1367    }
1368
1369    // -- Cross-site WebSocket --
1370
1371    #[tokio::test]
1372    async fn ws_upgrade_from_foreign_origin_is_rejected() {
1373        let req = HttpRequest::get("/graphql/ws")
1374            .header(axum::http::header::HOST, "127.0.0.1:4000")
1375            .header(axum::http::header::ORIGIN, "https://evil.com")
1376            .body(Body::empty())
1377            .unwrap();
1378        assert_eq!(run_browser(req).await, StatusCode::FORBIDDEN);
1379    }
1380
1381    #[tokio::test]
1382    async fn ws_upgrade_without_origin_is_allowed() {
1383        let req = HttpRequest::get("/graphql/ws")
1384            .header(axum::http::header::HOST, "127.0.0.1:4000")
1385            .body(Body::empty())
1386            .unwrap();
1387        assert_eq!(run_browser(req).await, StatusCode::OK);
1388    }
1389
1390    #[tokio::test]
1391    async fn configured_and_same_origin_are_allowed() {
1392        let req = HttpRequest::get("/graphql/ws")
1393            .header(axum::http::header::HOST, "127.0.0.1:4000")
1394            .header(axum::http::header::ORIGIN, "https://music.example.com")
1395            .body(Body::empty())
1396            .unwrap();
1397        assert_eq!(run_browser(req).await, StatusCode::OK);
1398
1399        // The bundled playground posts to the host it was served from.
1400        let req = json_post("/graphql")
1401            .header(axum::http::header::HOST, "127.0.0.1:4000")
1402            .header(axum::http::header::ORIGIN, "http://127.0.0.1:4000")
1403            .body(Body::empty())
1404            .unwrap();
1405        assert_eq!(run_browser(req).await, StatusCode::OK);
1406    }
1407
1408    // -- CSRF via a CORS-safelisted content type --
1409
1410    #[tokio::test]
1411    async fn text_plain_post_is_rejected() {
1412        let req = HttpRequest::post("/graphql")
1413            .header(axum::http::header::CONTENT_TYPE, "text/plain")
1414            .body(Body::from(r#"{"query":"mutation{clearQueue{ok}}"}"#))
1415            .unwrap();
1416        assert_eq!(run_browser(req).await, StatusCode::UNSUPPORTED_MEDIA_TYPE);
1417    }
1418
1419    #[tokio::test]
1420    async fn post_without_content_type_is_rejected() {
1421        let req = HttpRequest::post("/graphql").body(Body::empty()).unwrap();
1422        assert_eq!(run_browser(req).await, StatusCode::UNSUPPORTED_MEDIA_TYPE);
1423    }
1424
1425    // -- Load perimeter --
1426
1427    #[tokio::test]
1428    async fn load_perimeter_refuses_an_oversized_query_body() {
1429        async fn parse(_: async_graphql_axum::GraphQLRequest) -> StatusCode {
1430            StatusCode::OK
1431        }
1432        let app = load_perimeter(axum::Router::new().route("/graphql", post(parse)));
1433        // A valid query padded with whitespace, streamed so no Content-Length
1434        // warns the limit ahead of the bytes.
1435        let body = |padding: usize| {
1436            let chunks = [
1437                axum::body::Bytes::from_static(br#"{"query":"{__typename}""#),
1438                axum::body::Bytes::from(vec![b' '; padding]),
1439                axum::body::Bytes::from_static(b"}"),
1440            ];
1441            Body::from_stream(tokio_stream::iter(chunks.map(Ok::<_, std::io::Error>)))
1442        };
1443        let req = json_post("/graphql").body(body(1 << 10)).unwrap();
1444        assert_eq!(
1445            app.clone().oneshot(req).await.unwrap().status(),
1446            StatusCode::OK
1447        );
1448        let req = json_post("/graphql").body(body(3 << 20)).unwrap();
1449        assert_ne!(app.oneshot(req).await.unwrap().status(), StatusCode::OK);
1450    }
1451
1452    #[tokio::test]
1453    async fn load_perimeter_passes_requests_and_turns_panics_into_500s() {
1454        async fn boom() -> &'static str {
1455            panic!("resolver exploded");
1456        }
1457
1458        let app = load_perimeter(
1459            axum::Router::new()
1460                .route("/graphql", post(ok))
1461                .route("/boom", post(boom)),
1462        );
1463
1464        let req = json_post("/graphql").body(Body::empty()).unwrap();
1465        assert_eq!(
1466            app.clone().oneshot(req).await.unwrap().status(),
1467            StatusCode::OK
1468        );
1469
1470        // Without CatchPanicLayer this drops the connection with nothing logged.
1471        let req = json_post("/boom").body(Body::empty()).unwrap();
1472        assert_eq!(
1473            app.oneshot(req).await.unwrap().status(),
1474            StatusCode::INTERNAL_SERVER_ERROR
1475        );
1476    }
1477
1478    #[tokio::test]
1479    async fn json_post_is_accepted() {
1480        let req = json_post("/graphql").body(Body::empty()).unwrap();
1481        assert_eq!(run_browser(req).await, StatusCode::OK);
1482
1483        let req = HttpRequest::post("/graphql")
1484            .header(
1485                axum::http::header::CONTENT_TYPE,
1486                "application/json; charset=utf-8",
1487            )
1488            .body(Body::empty())
1489            .unwrap();
1490        assert_eq!(run_browser(req).await, StatusCode::OK);
1491    }
1492
1493    /// A server in an app is stopped and started again within one process:
1494    /// a stop ends its links rather than waiting them out, and leaves nothing
1495    /// behind that the next start trips over.
1496    #[test]
1497    fn an_embedded_server_ends_its_links_on_stop_and_starts_again() {
1498        use koan_core::auth::Role;
1499        use koan_core::db::connection::Database;
1500        use koan_core::player::commands::CommandChannel;
1501
1502        let _own = koan_core::config::SWITCHING
1503            .lock()
1504            .unwrap_or_else(|e| e.into_inner());
1505        let dir = tempfile::tempdir().unwrap();
1506        koan_core::config::set_config_dir(dir.path());
1507        let db_path = dir.path().join("koan.db");
1508        let db = Database::open(&db_path).unwrap();
1509        koan_core::db::schema::create_tables(&db.conn).unwrap();
1510        koan_core::db::queries::auth::create_user(&db.conn, "sharer", "hunter22", Role::Admin)
1511            .unwrap();
1512
1513        let start = || {
1514            ApiServer::start(ApiServerOpts {
1515                state: SharedPlayerState::new(),
1516                cmd_tx: CommandChannel::new().tx,
1517                pool: Arc::new(Pool::new(db_path.clone())),
1518                port: Some(0),
1519                bind: Some(std::net::Ipv4Addr::LOCALHOST.into()),
1520                subsonic_port: None,
1521                playground: false,
1522                viz: None,
1523                profile: Profile::Embedded,
1524                hosts: Vec::new(),
1525                session: None,
1526            })
1527            .unwrap()
1528        };
1529        // Subsonic answers, though `[subsonic]` is off by default.
1530        let ping = |addr: std::net::SocketAddr| {
1531            let body = reqwest::blocking::get(format!(
1532                "http://{addr}/rest/ping?u=sharer&p=hunter22&v=1.16.1&c=test&f=json"
1533            ))
1534            .unwrap()
1535            .text()
1536            .unwrap();
1537            assert!(body.contains(r#""status":"ok""#), "{body}");
1538        };
1539        let linked = || {
1540            crate::clients::registry()
1541                .list(Some("sharer"))
1542                .iter()
1543                .any(|c| c.device == "embedded-restart")
1544        };
1545
1546        let server = start();
1547        let addr = server.addr();
1548        ping(addr);
1549        let (mut socket, _) = tungstenite::connect(format!(
1550            "ws://{addr}/rest/koanLink?u=sharer&p=hunter22&v=1.16.1&c=test&device=embedded-restart"
1551        ))
1552        .unwrap();
1553        let deadline = std::time::Instant::now() + std::time::Duration::from_secs(10);
1554        while !linked() {
1555            assert!(
1556                std::time::Instant::now() < deadline,
1557                "the device never linked"
1558            );
1559            std::thread::sleep(std::time::Duration::from_millis(20));
1560        }
1561
1562        let stopping = std::time::Instant::now();
1563        server.stop();
1564        assert!(stopping.elapsed() < DRAIN, "stop waited out the drain");
1565        assert!(!linked(), "the link outlived the server");
1566        // The device is told, so it links again once the server is back.
1567        let closed = loop {
1568            match socket.read() {
1569                Ok(tungstenite::Message::Close(frame)) => break frame,
1570                Ok(_) => continue,
1571                Err(e) => panic!("no close frame: {e}"),
1572            }
1573        };
1574        assert_eq!(
1575            closed.map(|f| f.code),
1576            Some(tungstenite::protocol::frame::coding::CloseCode::Away)
1577        );
1578
1579        let again = start();
1580        ping(again.addr());
1581        again.stop();
1582    }
1583
1584    /// A port another process holds on loopback is taken, though the system
1585    /// would let a listener on every interface share it.
1586    #[test]
1587    fn an_embedded_server_refuses_a_port_held_on_loopback() {
1588        use koan_core::player::commands::CommandChannel;
1589
1590        let _own = koan_core::config::SWITCHING
1591            .lock()
1592            .unwrap_or_else(|e| e.into_inner());
1593        let dir = tempfile::tempdir().unwrap();
1594        koan_core::config::set_config_dir(dir.path());
1595        let held = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
1596        let port = held.local_addr().unwrap().port();
1597        let started = ApiServer::start(ApiServerOpts {
1598            state: SharedPlayerState::new(),
1599            cmd_tx: CommandChannel::new().tx,
1600            pool: Arc::new(Pool::new(dir.path().join("koan.db"))),
1601            port: Some(port),
1602            bind: Some(std::net::Ipv4Addr::UNSPECIFIED.into()),
1603            subsonic_port: None,
1604            playground: false,
1605            viz: None,
1606            profile: Profile::Embedded,
1607            hosts: Vec::new(),
1608            session: None,
1609        });
1610        let Err(why) = started else {
1611            panic!("started beside another listener on port {port}");
1612        };
1613        assert!(why.contains("already in use"), "{why}");
1614    }
1615}